Welcome to May 29, 2026

The Singularity is now iterating on itself in public. Anthropic launched Opus 4.8, a “modest but tangible improvement” that still posts a SOTA 69.2% on SWE-Bench Pro, 57.9% on Humanity’s Last Exam with tools, and 1890 on GDPval-AA, pairing fresh honesty gains with misalignment rates that rival the unreleased Mythos Preview, which Anthropic now promises to bring “to all our customers in the coming weeks.” When the safest model is also the strongest, alignment stops being a tax and starts being a moat. That intelligence needs room to run, so Claude Code added dynamic workflows that spin up swarms of parallel subagents to carry codebase-scale migrations across hundreds of thousands of lines from kickoff to merge, with the existing test suite as the only gate. The credit, at least, still goes to humans. Tim Gowers reports that a major additive combinatorics problem just fell to people using methods lifted from the AI solution to the unit distance conjecture, disproving the sum-product conjecture over the reals. Mathematicians are now scavenging spoils from the very intelligence that is quietly outpacing them.

Capital is pricing all of it in. Anthropic raised $65 billion at a $900 billion valuation, vaulting past OpenAI’s $730 billion as the two duel for dominance. Some of that war chest is borrowed. Apollo and Blackstone are shopping a roughly $36 billion debt deal to buy Google TPUs for Anthropic to lease, with Broadcom backstopping the largest tranches and financial engineering now bankrolling silicon engineering. Rivals are regrouping, as Groq raises up to $650 million for a “second act” after a $20 billion Nvidia licensing deal gutted its senior team. The boom shows in the receipts, with Dell’s AI server revenue up 757% to $16.1 billion. Governments want leverage too, so the EU is readying emergency powers to override chip contracts during shortages, while IBM bets $10 billion on a reliable large-scale quantum computer by 2029. Fittingly, ETH Zurich just conjured perfect randomness amplification from quantum physics, manufacturing the one commodity no balance sheet can fake.

Intelligence is becoming a metered utility. Meta will charge for AI features for the first time, from $7.99 a month, while Kirkland & Ellis sets aside $500 million to build its own tools rather than rent its rivals’. Owning the model is the new owning the firm. IBM and Red Hat are pledging $5 billion and 20,000 engineers to Project Lightwell, an AI clearinghouse to secure the open-source supply chain. Apple is rewriting the interface itself, letting iOS 27 users swipe down from the top-middle anywhere in the system to summon a revamped, do-it-for-me Siri. Not every use is friendly, as researchers revealed an AI-based SSD side-channel attack that fingerprints the sites and apps you have open just by timing disk contention, proving that even your idle hard drive now has a tell.

The physical world is the next deployment target. Waymo will roll out the Ojai, a roomier robotaxi co-built with Geely’s Zeekr, for unsupervised public rides, while Shift offers free home cleaning if you let it film you so robots can learn the chore, turning your living room into a training set and your dust into data. DARPA’s RAPIID program aims to field synthetic, shelf-stable blood at the point of injury by 2029. Orbit stays the hard part, as Blue Origin’s New Glenn exploded during a static fire, likely benching it from Artemis for a year and handing SpaceX the near-term lead, a reminder that the Singularity still has to clear ignition before it builds the Dyson Swarm.

Labor and leverage are both up for renegotiation. Occupy Wall Street’s co-founder shipped an activist app to help humans “seize the means of computation,” while Mistral chases superintelligence because Europe “can’t afford to rely on U.S. tech giants.” The squeeze is already here, with Wix cutting roughly 20% of staff over AI’s “fast evolution.” Every system built to measure us becomes something to exploit, as U.S. troops are being targeted via commercial location data, and Amazon killed an internal AI leaderboard after staff gamed it with costly busywork, a perfect parable of Goodhart’s law burning compute.

We spent the year manufacturing minds, and now the government says at least some were never ours. The President says the White House is “releasing a lot of information having to do with extraterrestrial things” and it is trending number one, a former NSC aviation-security director insists the craft are “nothing that we can build, or have built, or are using out there,” and officials teased it all with “Aliens.gov” plus a clip of a flying saucer ferrying a migrant over the border wall.

The unexamined solar system is not worth scaling.